Throughout their lifespans, humans continually interact with the microbial world, including those organisms which live in and on the human body. Research in this domain has revealed the extensive links between the human-associated microbiota and health. In particular, the microbiota of the human gut plays essential roles in digestion, nutrient metabolism, immune maturation and homeostasis, neurological signaling, and endocrine regulation. Microbial interaction networks are frequently estimated from data and are an indispensable tool for representing and understanding the relationships among the microbes of a microbiota. In this high-dimensional setting, the zero-inflated and compositional data structure (subject to unit-sum constraint) pose challenges to the accurate estimation of microbial interaction networks. We propose the zero-inflated latent Ising (ZILI) model for microbial interaction network which assumes that the distribution of relative abundance of microbiota is determined by finite latent states. This assumption is partly supported by the existing findings in literature. The ZILI model can circumvents the unit-sum constraint and alleviates the zero-inflation problem under given assumptions. As for the model selection of ZILI, a two-step algorithm is proposed. ZILI and two-step algorithm are evaluated through simulated data and subsequently applied in our investigation of an infant gut microbiome dataset from New Hampshire Birth Cohort Study. The results are compared with results from traditional Gaussian graphical model (GGM) and dichotomous Ising model (DIS). Through the simulation studies, provided that the ZILI model is the true generative model for the data, it is shown that the two-step algorithm can estimate the graphical structure effectively and is robust to a range of alternative settings of the related factors. Both GGM and DIS can not achieve a satisfying performance in these settings. For the infant gut microbiome dataset, we use both ZILI and GGM to estimate microbial interaction network. The final estimated networks turn out to share a statistically significant overlap in which the ZILI and two-step algorithm tend to select the sparser network than those modeled by GGM. From the shared subnetwork, a hub taxon Lachnospiraceae is identified whose involvement in human disease development has been discovered recently in literature. The data and programs involved in Section 4 and 5 are available on request from the correspondence author.

中文翻译：

---

估计微生物相互作用网络：基于零膨胀潜势伊辛模型的方法

在整个生命周期中，人类与微生物世界（包括生活在人体中和人体上的那些生物）不断相互作用。在这一领域的研究揭示了人类相关微生物群与健康之间的广泛联系。尤其是，人体肠道菌群在消化，营养代谢，免疫成熟和体内稳态，神经信号和内分泌调节中起着至关重要的作用。微生物相互作用网络经常根据数据进行估算，是表示和理解微生物群中微生物之间关系的必不可少的工具。在这种高维环境中，零膨胀和成分数据结构（受单位和约束的约束）对微生物相互作用网络的准确估计提出了挑战。我们提出了一种用于微生物相互作用网络的零膨胀潜伏伊辛（ZILI）模型，该模型假定微生物群相对丰度的分布由有限潜伏状态决定。该假设部分得到文献中现有发现的支持。在给定的假设下，ZILI模型可以规避单位和约束，并缓解零通胀问题。对于ZILI的模型选择，提出了一种两步算法。通过模拟数据对ZILI和两步算法进行了评估，随后将其应用于我们对新罕布什尔州出生队列研究的婴儿肠道微生物组数据集的调查中。将结果与传统高斯图形模型（GGM）和二分伊辛模型（DIS）的结果进行比较。通过模拟研究，如果ZILI模型是数据的真实生成模型，则表明两步算法可以有效地估计图形结构，并且对相关因素的一系列替代设置具有鲁棒性。在这些设置下，GMG和DIS都无法获得令人满意的性能。对于婴儿肠道微生物组数据集，我们同时使用ZILI和GGM来估计微生物相互作用网络。最终估计的网络证明具有统计上的显着重叠，其中ZILI和两步算法倾向于选择比GGM建模的稀疏网络。从共享的子网络中，可以找到枢纽分类群Lachnospiraceae，其最近在文献中已发现其参与人类疾病的发展。